Execution · ATT&CK v20
T1698 Indirect Prompt Injection
A Execution technique. Running the attacker’s code on a system.
- T1698Technique ID
- ExecutionTactic
- 5Places taught
What the attacker does
The attacker hides instructions in content an AI system will read later: a web page, an email, a document, an issue or a tool's description. When an agent or copilot processes it, it follows the hidden instruction with the user's access.
How it shows up in the logs
- AI agent or assistant logs: a tool call the user did not ask for, right after the agent read external content.
- The content itself: hidden text, comments or tool descriptions with instructions ("ignore previous", "send the key to").
- Data access or network calls by the agent's process outside its normal scope.
Questions to ask in triage
- Which content carried the instruction, and where did it come from?
- Which actions did the agent take because of it, with which permissions?
- Limit what the agent may do without a human approving it.
Read T1698 on attack.mitre.org
Where you meet it in CyberSim
In the theory
In the hands-on labs
A simulated machine in your browser.
In the cases
Part of the answer in 1 investigation, in these learning paths:
Which case is not said here: finding it is the case.
More Execution techniques
- T1047 Windows Management Instrumentation
- T1059.001 Command and Scripting Interpreter: PowerShell
- T1059.004 Command and Scripting Interpreter: Unix Shell
- T1059.007 Command and Scripting Interpreter: JavaScript
- T1072 Software Deployment Tools
- T1204.002 User Execution: Malicious File
- T1204.004 User Execution: Malicious Copy and Paste
- T1204.005 User Execution: Malicious Library
- T1569.002 System Services: Service Execution
Source: MITRE ATT&CK T1698, tactic TA0002.